haarmatrix

Haarmatrix, in the context of signal processing and wavelet theory, refers to the matrix representation of the discrete Haar wavelet transform. It is typically defined for lengths n that are powers of two (n = 2^m) and forms an orthogonal (or unitary, in the complex case) matrix whose rows correspond to a Haar wavelet basis of length n. The first row represents the scaling function (the average component), while the remaining rows correspond to detail components at increasingly finer dyadic scales.

Construction and interpretation: The Haar matrix is built from a set of piecewise constant basis vectors. The scaling row is a constant vector with all entries equal, normalized by 1/√n. The detail rows are generated by taking blocks of length 2^k, where within each block the vector has +1 on the first half and -1 on the second half, and zeros outside the block, with appropriate normalization. As k runs from 1 to m and across all block positions, these vectors form an orthonormal basis, and arranging them as rows yields the Haar matrix.

Properties: The Haar matrix is real and orthogonal, so H^T H = I and H^{-1} = H^T. It provides a fast, multiscale decomposition of a signal into averages and differences across scales, with a computational complexity of O(n log n) using a fast Haar transform algorithm. Normalization conventions can vary, but the first row typically contains equal-weight entries and subsequent rows are scaled to maintain unit norm.

Applications and variants: It is used for data compression, denoising, feature extraction, and time–frequency analysis, similar to other wavelet transforms but with a very simple, interval-based structure. Variants exist for non-power-of-two lengths and different normalization schemes, though the core idea remains the multiscale, piecewise-constant basis.